Machine Learning Classification of Palatal Salivary Gland Tumors Using Clinical and Demographic Descriptors

Area of Science:

• Oncology
• Medical Informatics
• Computational Biology

Background:

• Palatal salivary gland tumors represent a diverse group of neoplasms.
• Accurate clinical classification is crucial for appropriate patient management and treatment planning.

Purpose of the Study:

• To evaluate the efficacy of machine learning (ML) models in classifying palatal salivary gland tumors.
• To assess the utility of demographic and clinical data for tumor classification using ML.

Main Methods:

• Four ML models (XGBoost, MLP, SVM, RF) were trained and tested on data from 100 patients.
• The study employed hyperparameter optimization (Grid Search) and fivefold cross-validation.
• Performance was measured using accuracy, macro-average sensitivity, specificity, precision, and F1-score.

Main Results:

• XGBoost and MLP achieved the highest mean accuracy (81%), followed by SVM (80%) and RF (79%).
• All models exhibited high specificity (85%-90%) but low macro-averaged sensitivity and F1-scores (<75%).
• Model performance was excellent for pleomorphic adenoma (PA) but decreased for rarer malignant tumors.

Conclusions:

• ML is a viable tool for palatal salivary gland tumor classification, offering high specificity.
• Limited sensitivity is attributed to the underrepresentation of rare malignant tumor subclasses.
• XGBoost emerged as the most robust and computationally efficient model.